Research On And Implementation Of Precise Terrain Classification For Polarimetric SAR Images

Polarimetric synthetic aperture radar(PolSAR)is one of the most advanced remote sensors in recent years.As an essential part and a key technology for PolSAR image interpretation, PolSAR image classification has been playing an important role in many fields of both civil and military applications. According to some problems of PolSAR image classification methods in recent years, we primarily study on the classification technology from three aspects:spatial relationship, spatial complexity adaptive and number-of-classes adaptive. The main work and contributions accomplished in this paper are as follows:1) We take the spatial relations between pixels into consideration and introduce the concept of superpixel in the field of computer vision. With good use of the inherent statistical characteristics and contour information of PolSAR data, we present a novel superpixel-based PolSAR image classification method using Normalized Cut. The classification results are very clear and easy to understand.2) Incorporating H/α-Wishart clustering, quad-tree decomposition and Wishart markov random field theory, we present a complexity adaptive segmentation method for PolSAR Images. The method integrates spatial adaptivity and the experimental results show that it can keep information of the details in PolSAR images.3) We first introduce a tool in the field of knowledge and data engineering, which is Visual Assessment of Cluster Tendency. Then incorporating superpixel generation method, we present a novel superpixel -based classification framework with an adaptive number of classes for PolSAR images. Although without any guidance of prior knowledge, this method can effectively estimate the number of classes and each class center in the image. Then we can use these for unsupervised classification of PolSAR images. This framework is capable of improving the classification accuracy, making the results more understandable and easier for further analysis. Additionally, we apply this framework for high-resolution SAR images. Combined with analysis of gray-scale and texture features, we also make it work well for high-resolution SAR images. The experiment result shows that the proposed method provides a promising performance for high-resolution SAR image classification.